Make pipeline able to load processor

src/transformers/pipelines/__init__.py

@@ -28,7 +28,9 @@
 from ..models.auto.feature_extraction_auto import FEATURE_EXTRACTOR_MAPPING, AutoFeatureExtractor
 from ..models.auto.image_processing_auto import IMAGE_PROCESSOR_MAPPING, AutoImageProcessor
 from ..models.auto.modeling_auto import AutoModelForDepthEstimation, AutoModelForImageToImage
+from ..models.auto.processing_auto import PROCESSOR_MAPPING, AutoProcessor
 from ..models.auto.tokenization_auto import TOKENIZER_MAPPING, AutoTokenizer
+from ..processing_utils import ProcessorMixin
 from ..tokenization_utils import PreTrainedTokenizer
 from ..utils import (
     CONFIG_NAME,
@@ -556,6 +558,7 @@ def pipeline(
     tokenizer: Optional[Union[str, PreTrainedTokenizer, "PreTrainedTokenizerFast"]] = None,
     feature_extractor: Optional[Union[str, PreTrainedFeatureExtractor]] = None,
     image_processor: Optional[Union[str, BaseImageProcessor]] = None,
+    processor: Optional[Union[str, ProcessorMixin]] = None,
     framework: Optional[str] = None,
     revision: Optional[str] = None,
     use_fast: bool = True,
@@ -644,6 +647,25 @@ def pipeline(
             `model` is not specified or not a string, then the default feature extractor for `config` is loaded (if it
             is a string). However, if `config` is also not given or not a string, then the default feature extractor
             for the given `task` will be loaded.
+        image_procesor (`str` or [`BaseImageProcessor`], *optional*):
+            The image processor that will be used by the pipeline to preprocess images for the model. This can be a
+            model identifier or an actual image processor inheriting from [`BaseImageProcessor`].
+
+            Image processors are used for Vision models and multi-modal models that require image inputs. Multi-modal
+            models will also require a tokenizer to be passed.
+
+            If not provided, the default image processor for the given `model` will be loaded (if it is a string). If
+            `model` is not specified or not a string, then the default image processor for `config` is loaded (if it is
+            a string).
+        processor (`str` or [`ProcessorMixin`], *optional*):
+            The processor that will be used by the pipeline to preprocess data for the model. This can be a model
+            identifier or an actual processor inheriting from [`ProcessorMixin`].
+
+            Processors are used for multi-modal models that require multi-modal inputs, for example, a model that
+            requires both text and image inputs.
+
+            If not provided, the default processor for the given `model` will be loaded (if it is a string). If `model`
+            is not specified or not a string, then the default processor for `config` is loaded (if it is a string).
         framework (`str`, *optional*):
             The framework to use, either `"pt"` for PyTorch or `"tf"` for TensorFlow. The specified framework must be
             installed.
@@ -905,13 +927,29 @@ def pipeline(
 
     model_config = model.config
     hub_kwargs["_commit_hash"] = model.config._commit_hash
+
     load_tokenizer = (
         type(model_config) in TOKENIZER_MAPPING
         or model_config.tokenizer_class is not None
         or isinstance(tokenizer, str)
     )
-    load_feature_extractor = type(model_config) in FEATURE_EXTRACTOR_MAPPING or feature_extractor is not None
-    load_image_processor = type(model_config) in IMAGE_PROCESSOR_MAPPING or image_processor is not None
+    load_feature_extractor = (
+        type(model_config) in FEATURE_EXTRACTOR_MAPPING
+        or feature_extractor is not None
+        or isinstance(feature_extractor, str)
+    )
+    load_image_processor = (
+        type(model_config) in IMAGE_PROCESSOR_MAPPING
+        or image_processor is not None
+        or isinstance(image_processor, str)
+    )
+    load_processor = type(model_config) in PROCESSOR_MAPPING or processor is not None or isinstance(processor, str)
+
+    # Check that pipeline class required loading
+    load_tokenizer = load_tokenizer and pipeline_class._load_tokenizer
+    load_feature_extractor = load_feature_extractor and pipeline_class._load_feature_extractor
+    load_image_processor = load_image_processor and pipeline_class._load_image_processor
+    load_processor = load_processor and pipeline_class._load_processor
 
     # If `model` (instance of `PretrainedModel` instead of `str`) is passed (and/or same for config), while
     # `image_processor` or `feature_extractor` is `None`, the loading will fail. This happens particularly for some
@@ -1074,6 +1112,32 @@ def pipeline(
                     if not is_pyctcdecode_available():
                         logger.warning("Try to install `pyctcdecode`: `pip install pyctcdecode")
 
+    if load_processor:
+        # Try to infer processor from model or config name (if provided as str)
+        if processor is None:
+            if isinstance(model_name, str):
+                processor = model_name
+            elif isinstance(config, str):
+                processor = config
+            else:
+                # Impossible to guess what is the right processor here
+                raise Exception(
+                    "Impossible to guess which processor to use. "
+                    "Please provide a processor instance or a path/identifier "
+                    "to a processor."
+                )
+
+        # Instantiate processor if needed
+        if isinstance(processor, (str, tuple)):
+            processor = AutoProcessor.from_pretrained(processor, _from_pipeline=task, **hub_kwargs, **model_kwargs)
+            if not isinstance(processor, ProcessorMixin):
+                warnings.warn(
+                    f"Processor will be not loaded, because {processor} is not an instance of `ProcessorMixin`. "
+                    f"Got type `{type(processor)}` instead.",
+                    UserWarning,
+                )
+                processor = None
+
     if task == "translation" and model.config.task_specific_params:
         for key in model.config.task_specific_params:
             if key.startswith("translation"):
@@ -1099,4 +1163,7 @@ def pipeline(
     if device is not None:
         kwargs["device"] = device
 
+    if processor is not None:
+        kwargs["processor"] = processor
+
     return pipeline_class(model=model, framework=framework, task=task, **kwargs)

src/transformers/pipelines/base.py

@@ -34,6 +34,7 @@
 from ..image_processing_utils import BaseImageProcessor
 from ..modelcard import ModelCard
 from ..models.auto.configuration_auto import AutoConfig
+from ..processing_utils import ProcessorMixin
 from ..tokenization_utils import PreTrainedTokenizer
 from ..utils import (
     ModelOutput,
@@ -716,6 +717,7 @@ def build_pipeline_init_args(
     has_tokenizer: bool = False,
     has_feature_extractor: bool = False,
     has_image_processor: bool = False,
+    has_processor: bool = False,
     supports_binary_output: bool = True,
 ) -> str:
     docstring = r"""
@@ -738,6 +740,11 @@ def build_pipeline_init_args(
         image_processor ([`BaseImageProcessor`]):
             The image processor that will be used by the pipeline to encode data for the model. This object inherits from
             [`BaseImageProcessor`]."""
+    if has_processor:
+        docstring += r"""
+        processor ([`ProcessorMixin`]):
+            The processor that will be used by the pipeline to encode data for the model. This object inherits from
+            [`ProcessorMixin`]."""
     docstring += r"""
         modelcard (`str` or [`ModelCard`], *optional*):
             Model card attributed to the model for this pipeline.
@@ -774,7 +781,11 @@ def build_pipeline_init_args(
 
 
 PIPELINE_INIT_ARGS = build_pipeline_init_args(
-    has_tokenizer=True, has_feature_extractor=True, has_image_processor=True, supports_binary_output=True
+    has_tokenizer=True,
+    has_feature_extractor=True,
+    has_image_processor=True,
+    has_processor=True,
+    supports_binary_output=True,
 )
 
 
@@ -805,6 +816,18 @@ class Pipeline(_ScikitCompat, PushToHubMixin):
     constructor argument. If set to `True`, the output will be stored in the pickle format.
     """
 
+    # Previously, pipelines support only `tokenizer`, `feature_extractor`, and `image_processor`.
+    # As we start adding `processor`, we want to avoid loading processor for some pipelines, that don't required it,
+    # because, for example, use `image_processor` and `tokenizer` separately.
+    # However, we want to enable it for new pipelines. Moreover, this allow us to granularly control loading components
+    # and avoid loading tokenizer/image_processor/feature_extractor twice: once as a separate object
+    # and once in the processor. The following flags a set this way for backward compatibility ans might be overridden
+    # in specific Pipeline class.
+    _load_processor = False
+    _load_image_processor = True
+    _load_feature_extractor = True
+    _load_tokenizer = True
+
     default_input_names = None
 
     def __init__(
@@ -813,6 +836,7 @@ def __init__(
         tokenizer: Optional[PreTrainedTokenizer] = None,
         feature_extractor: Optional[PreTrainedFeatureExtractor] = None,
         image_processor: Optional[BaseImageProcessor] = None,
+        processor: Optional[ProcessorMixin] = None,
         modelcard: Optional[ModelCard] = None,
         framework: Optional[str] = None,
         task: str = "",
@@ -830,6 +854,7 @@ def __init__(
         self.tokenizer = tokenizer
         self.feature_extractor = feature_extractor
         self.image_processor = image_processor
+        self.processor = processor
         self.modelcard = modelcard
         self.framework = framework
 

tests/models/altclip/test_modeling_altclip.py

@@ -436,9 +436,16 @@ class AltCLIPModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.TestCase)
 
     # TODO: Fix the failed tests when this model gets more usage
     def is_pipeline_test_to_skip(
-        self, pipeline_test_casse_name, config_class, model_architecture, tokenizer_name, processor_name
+        self,
+        pipeline_test_case_name,
+        config_class,
+        model_architecture,
+        tokenizer_name,
+        image_processor_name,
+        feature_extractor_name,
+        processor_name,
     ):
-        if pipeline_test_casse_name == "FeatureExtractionPipelineTests":
+        if pipeline_test_case_name == "FeatureExtractionPipelineTests":
             return True
 
         return False

tests/models/audio_spectrogram_transformer/test_modeling_audio_spectrogram_transformer.py

@@ -167,9 +167,16 @@ class ASTModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.TestCase):
 
     # TODO: Fix the failed tests when this model gets more usage
     def is_pipeline_test_to_skip(
-        self, pipeline_test_casse_name, config_class, model_architecture, tokenizer_name, processor_name
+        self,
+        pipeline_test_case_name,
+        config_class,
+        model_architecture,
+        tokenizer_name,
+        image_processor_name,
+        feature_extractor_name,
+        processor_name,
     ):
-        if pipeline_test_casse_name == "AudioClassificationPipelineTests":
+        if pipeline_test_case_name == "AudioClassificationPipelineTests":
             return True
 
         return False

tests/models/bigbird_pegasus/test_modeling_bigbird_pegasus.py

@@ -276,9 +276,16 @@ class BigBirdPegasusModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineT
 
     # TODO: Fix the failed tests
     def is_pipeline_test_to_skip(
-        self, pipeline_test_casse_name, config_class, model_architecture, tokenizer_name, processor_name
+        self,
+        pipeline_test_case_name,
+        config_class,
+        model_architecture,
+        tokenizer_name,
+        image_processor_name,
+        feature_extractor_name,
+        processor_name,
     ):
-        if pipeline_test_casse_name == "QAPipelineTests" and not tokenizer_name.endswith("Fast"):
+        if pipeline_test_case_name == "QAPipelineTests" and not tokenizer_name.endswith("Fast"):
             return True
 
         return False

tests/models/blenderbot_small/test_modeling_blenderbot_small.py

@@ -236,9 +236,16 @@ class BlenderbotSmallModelTest(ModelTesterMixin, GenerationTesterMixin, Pipeline
 
     # TODO: Fix the failed tests when this model gets more usage
     def is_pipeline_test_to_skip(
-        self, pipeline_test_casse_name, config_class, model_architecture, tokenizer_name, processor_name
+        self,
+        pipeline_test_case_name,
+        config_class,
+        model_architecture,
+        tokenizer_name,
+        image_processor_name,
+        feature_extractor_name,
+        processor_name,
     ):
-        return pipeline_test_casse_name == "TextGenerationPipelineTests"
+        return pipeline_test_case_name == "TextGenerationPipelineTests"
 
     def setUp(self):
         self.model_tester = BlenderbotSmallModelTester(self)

tests/models/blenderbot_small/test_modeling_flax_blenderbot_small.py

@@ -321,9 +321,16 @@ class FlaxBlenderbotSmallModelTest(FlaxModelTesterMixin, unittest.TestCase, Flax
     all_generative_model_classes = (FlaxBlenderbotSmallForConditionalGeneration,) if is_flax_available() else ()
 
     def is_pipeline_test_to_skip(
-        self, pipeline_test_casse_name, config_class, model_architecture, tokenizer_name, processor_name
+        self,
+        pipeline_test_case_name,
+        config_class,
+        model_architecture,
+        tokenizer_name,
+        image_processor_name,
+        feature_extractor_name,
+        processor_name,
     ):
-        return pipeline_test_casse_name == "TextGenerationPipelineTests"
+        return pipeline_test_case_name == "TextGenerationPipelineTests"
 
     def setUp(self):
         self.model_tester = FlaxBlenderbotSmallModelTester(self)

tests/models/blenderbot_small/test_modeling_tf_blenderbot_small.py

@@ -198,9 +198,16 @@ class TFBlenderbotSmallModelTest(TFModelTesterMixin, PipelineTesterMixin, unitte
     test_onnx = False
 
     def is_pipeline_test_to_skip(
-        self, pipeline_test_casse_name, config_class, model_architecture, tokenizer_name, processor_name
+        self,
+        pipeline_test_case_name,
+        config_class,
+        model_architecture,
+        tokenizer_name,
+        image_processor_name,
+        feature_extractor_name,
+        processor_name,
     ):
-        return pipeline_test_casse_name == "TextGenerationPipelineTests"
+        return pipeline_test_case_name == "TextGenerationPipelineTests"
 
     def setUp(self):
         self.model_tester = TFBlenderbotSmallModelTester(self)

tests/models/bros/test_modeling_bros.py

@@ -295,7 +295,14 @@ class BrosModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.TestCase):
     # BROS requires `bbox` in the inputs which doesn't fit into the above 2 pipelines' input formats.
     # see https://github.com/huggingface/transformers/pull/26294
     def is_pipeline_test_to_skip(
-        self, pipeline_test_casse_name, config_class, model_architecture, tokenizer_name, processor_name
+        self,
+        pipeline_test_case_name,
+        config_class,
+        model_architecture,
+        tokenizer_name,
+        image_processor_name,
+        feature_extractor_name,
+        processor_name,
     ):
         return True
 

tests/models/cpm/test_tokenization_cpm.py

@@ -22,7 +22,14 @@
 class CpmTokenizationTest(unittest.TestCase):
     # There is no `CpmModel`
     def is_pipeline_test_to_skip(
-        self, pipeline_test_casse_name, config_class, model_architecture, tokenizer_name, processor_name
+        self,
+        pipeline_test_case_name,
+        config_class,
+        model_architecture,
+        tokenizer_name,
+        image_processor_name,
+        feature_extractor_name,
+        processor_name,
     ):
         return True
 

tests/models/ctrl/test_modeling_ctrl.py

@@ -211,9 +211,16 @@ class CTRLModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTesterMixin
 
     # TODO: Fix the failed tests
     def is_pipeline_test_to_skip(
-        self, pipeline_test_casse_name, config_class, model_architecture, tokenizer_name, processor_name
+        self,
+        pipeline_test_case_name,
+        config_class,
+        model_architecture,
+        tokenizer_name,
+        image_processor_name,
+        feature_extractor_name,
+        processor_name,
     ):
-        if pipeline_test_casse_name == "ZeroShotClassificationPipelineTests":
+        if pipeline_test_case_name == "ZeroShotClassificationPipelineTests":
             # Get `tokenizer does not have a padding token` error for both fast/slow tokenizers.
             # `CTRLConfig` was never used in pipeline tests, either because of a missing checkpoint or because a tiny
             # config could not be created.

tests/models/ctrl/test_modeling_tf_ctrl.py

@@ -189,9 +189,16 @@ class TFCTRLModelTest(TFModelTesterMixin, PipelineTesterMixin, unittest.TestCase
 
     # TODO: Fix the failed tests
     def is_pipeline_test_to_skip(
-        self, pipeline_test_casse_name, config_class, model_architecture, tokenizer_name, processor_name
+        self,
+        pipeline_test_case_name,
+        config_class,
+        model_architecture,
+        tokenizer_name,
+        image_processor_name,
+        feature_extractor_name,
+        processor_name,
     ):
-        if pipeline_test_casse_name == "ZeroShotClassificationPipelineTests":
+        if pipeline_test_case_name == "ZeroShotClassificationPipelineTests":
             # Get `tokenizer does not have a padding token` error for both fast/slow tokenizers.
             # `CTRLConfig` was never used in pipeline tests, either because of a missing checkpoint or because a tiny
             # config could not be created.

tests/models/falcon/test_modeling_falcon.py

@@ -312,7 +312,14 @@ class FalconModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTesterMix
 
     # TODO (ydshieh): Check this. See https://app.circleci.com/pipelines/github/huggingface/transformers/79245/workflows/9490ef58-79c2-410d-8f51-e3495156cf9c/jobs/1012146
     def is_pipeline_test_to_skip(
-        self, pipeline_test_casse_name, config_class, model_architecture, tokenizer_name, processor_name
+        self,
+        pipeline_test_case_name,
+        config_class,
+        model_architecture,
+        tokenizer_name,
+        image_processor_name,
+        feature_extractor_name,
+        processor_name,
     ):
         return True